bayesian statistical approach to dark matter direct detection
TRANSCRIPT
Chiara Arina
Bols
hoi s
imul
atio
n, N
ASA
Bayesian statistical approach to dark matter direct detection experiments
PASCOS 201319th International Symposium on Particles, Strings and Cosmology
Taipei, November 20-26, 2013
1
Outline
• Bayesian analysis of direct detection data motivated by(i) Tension between experiments (4 hints of detection and exclusion bounds)(ii) Experimental systematics (e.g. Leff, quenching factors) and backgrounds(iii) Astrophysical uncertainties in both the halo parameters and velocity distribution
• Bayesian Evidence for model comparison and compatibility• Best scenario that accommodates XENON100 and the hints of detection (DAMA, CoGeNT, CDMS-Si, CRESST)
• Best particle physics scenario for hints of detection
• Quantitative measure of incompatibility between XENON100 and hints of detection
• Conclusions
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 20132
• CA, J.Hamann and Y.Wong, JCAP 1109 (2011)• CA, J.Hamann, R.Trotta and Y.Wong, JCAP 1203 (2012)• CA, Phys.Rev.D86 (2012)• CA, arXiv: 1310.5718, invited review for special issue of PDU
Bayesian Inference framework
Likelihood(proper of each EXP)
PriorPosterior probabilityfunction (PDF)
data
�i
�k
theoretical model parametersnuisance parameters = astrophysics and systematics
� = {�1, ..., �n,�a, ...,�z}
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 20133
Bayesian Inference framework
Likelihood(proper of each EXP)
PriorPosterior probabilityfunction (PDF)
data
�i
�k
theoretical model parametersnuisance parameters = astrophysics and systematics
� = {�1, ..., �n,�a, ...,�z}
Common prior choices that do not favour any parameter region
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 20133
Bayesian Inference framework
Likelihood(proper of each EXP)
PriorPosterior probabilityfunction (PDF)
data
�i
�k
theoretical model parametersnuisance parameters = astrophysics and systematics
� = {�1, ..., �n,�a, ...,�z}
Posterior sampled with nested sampling techniques (MultiNest) given the likelihood and the priorand marginalized over nuisance parameters
Common prior choices that do not favour any parameter region
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 20133
Bayesian Inference framework
Likelihood(proper of each EXP)
PriorPosterior probabilityfunction (PDF)
data
�i
�k
theoretical model parametersnuisance parameters = astrophysics and systematics
� = {�1, ..., �n,�a, ...,�z}
Posterior sampled with nested sampling techniques (MultiNest) given the likelihood and the priorand marginalized over nuisance parameters
Common prior choices that do not favour any parameter region
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 20133
Profile Likelihood is prior independent (comparison with frequentist approach)
4 C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
Bayesian Inference frameworkMarginalization over all nuisance/new physics parameters
Background and systematics
Astrophysical parameters(common to all exp)
Beyond elastic SI scattering(common to all exp)
Inference for constraining data , example with DAMA1D marginalized posterior PDFquenching factors (nuisance)
Matches with profile likelihood analysisC. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 20135
log10(mDM) [GeV]
log 10
(σnSI
) [cm
2 ]
Posterior PDFElastic scattering
0 0.5 1 1.5 2 2.5 3−41.5
−41
−40.5
−40
−39.5
−39
2D marginal credible regions at 90 and 99% (shaded f(v)=NFW, solid blue f(v)= MB)
Inference for non constraining data: CDMS-Si
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 20136
log10(mDM) [GeV]
log 10
(σnSI
) [cm
2 ]
Posterior pdf
1 1.5 2 2.5 3−44
−43
−42
−41
−40
−39
−38
log10(mDM) [GeV]
log 10
(σnSI
) [cm
2 ]
Profile likelihood
1 1.5 2 2.5 3−44
−43
−42
−41
−40
−39
−38
2D marginal credible regions at 68 and 90%for fixed astrophysics
• Likelihood follows a Poisson distribution with spectral information • 3 events seen with estimated bckg of 0.7: not constraining data
data from CDMS-Si collaboration arXiv:1304.4279
Inference for exclusion bounds: XENON100
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 20137
Data are not constraining therefore the upper bound depends on the prior choice:
log10(mDM) [GeV]
log 10
(σnSI
) [cm
2 ]
Posterior pdfElastic Scattering
0.5 1 1.5 2 2.5 3−45
−44
−43
−42
−41
−40
−39
−38
Invariant exclusion bound based on the S signal with bayesian interpretation:
2D marginal credible regions at 90% +
from arXiv:1104.2549
- 2 events seen, likelihood follows a Poisson distribution - expected background of 1. +- 0.8, analytical marginalization - considered Poisson fluctuation below threshold
data from XENON100 collaboration, arXiv:1207.5988
Inference for elastic SI scattering
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 20138
log10(mDM) [GeV]
log 10
(σnSI
) [cm
2 ]
Posterior PDF Elastic scattering
Marginalized astro (NFW)
1 1.5 2 2.5 3−45
−44
−43
−42
−41
−40
−39
−38
- The marginalization over astrophysics does not improve the compatibility between XENON100 and all detection hints- The XENON100 bound is less stringent at masses larger than 30 GeV than the one of the collaboration because of
the approximate likelihood - Same analysis can be done with LUX, more difficult to reconcile low mass regions, as its threshold is at 2 PE
arXiv:1207.5988profile likelihood analysisfixed astrophysics
Inference for isospin violating scattering
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 20139
log10(mDM) [GeV]
log 10
(σnSI
) [cm
2 ]
Posterior PDFIsospin violating scatteringMarginalized astro (NFW)
1 1.5 2 2.5 3−44
−43
−42
−41
−40
−39
−38
−37
−36
- The extra parameter is not supported/constrained by current data
- The marginalization over the parameter causes a volume effect: detection regions becomes larger and the exclusion bound moves to the right
- Within the Bayesian approach the hint regions become compatible with the 90% CL of XENON100
- Inelastic and exothermic dark matter have same volume effect, however the agreement between detection regions and exclusion bounds is worst than isospin violating scenario
Assumption that interaction of WIMP with proton and neutron is of different strength:
Bayesian evidence for model comparison
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 201310
Bayesian evidence1. model averaged likelihood2. contains notion of Occam’s razor principle3. used for model comparison and statistical test
Posterior pdf for a model:
�(M0) = �(M1)(non committal prior)
Bayes factor: ratio of model’s evidences
Empirical Jeffreys’ scale
Combined fit and model comparison
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 201311
Combined fit pushes the quenching factor and the local standard at rest at corner values (1D pdf not flat anymore but peaked at qNa=0.6)
log10(mDM) [GeV]
log 10
(σnSI
) [cm
2 ]
Posterior PDF Elastic scattering
Marginalized astro (NFW)
0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4−41
−40.5
−40
−39.5
−39
−38.5
−38
• Evidence between elastic and isospin violating scenarios is inconclusive• Inelastic and exothermic moderately disfavored with respect to elastic SI
(In)Compatibility between XENON100 and detection hints ?
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 201312
Nevents
ln L
(Nev
ents |
D)
1 σ
2 σ
3 σ
Observed
0 10 20 30 40 50 60 70 80 90 100−20
−18
−16
−14
−12
−10
−8
−6
−4
−2
0Consider a data set which is given by two subsets:1. = this is the data set which is taken as true, hence fixed2. = data set that we want to test, that is we want to quantify if it is compatible with
Elastic scattering
Isospin violating framework: likelihood ratio in data space gives incompatibility at
Summary
• Bayesian approach for XENON100, DAMA, CoGeNT, CRESST and CDMSI-Si data with marginalization over the systematics and nuisance parameters characteristic of each experiment (can be applied to LUX, similar to XENON100 procedure)
• Inclusion of velocity distributions arising from DM density profile and marginalization over astrophysical variables (NFW)
• Difficult to reconcile at 90% CL all detection hints and XENON100
• Going beyond the elastic SI scattering (isospin violating, inelastic and exothermic scattering) ameliorates the compatibility between experiments: the additional physics parameter is not constrained by the current data
• Astrophysical uncertainties can not be yet constrained by direct detection experiment alone (however combined fit can constrain astrophysics)
• Combined fit implies large value of the quenching factor on Sodium for DAMA and small local standard of rest velocity
• For hints of detection the elastic and isospin violating scenarios have the strongest support form the data; isospin violating framework ameliorate the compatibility between hints of detection and exclusion bounds
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 201313
Inference for inelastic/exothermic SI scattering
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 201315
log10(mDM) [GeV]
log 10
(σnSI
) [cm
2 ]
Posterior PDFExothermic scattering
Marginalized astro (NFW)
0.5 1 1.5 2 2.5 3−44
−43.5
−43
−42.5
−42
−41.5
−41
−40.5
−40
−39.5
−39
log10(mDM) [GeV]
log 10
(σSI n
) [cm
2 ]
Posterior PDFInelastic scatteringMarginalized astro (NFW)
1 1.5 2 2.5 3−44
−43
−42
−41
−40
−39
−38
−37
−36
16
Changing the WIMP physics interaction Isospin violating interaction
• Assumption that interaction of WIMP with proton and neutron is of different strength:
• Defined a mean SI cross-section with an effective couplings to nuclei:
Feng et al. ’11
(Schwetz, Zupan ’11)
• Example of realization in WIMPs model: The couplings neutralino-squark-quark violate isospin, however in the most common scenarios they are not the dominant contributions to elastic scattering
• Other possibilities: long range interactions, inelastic scattering, spin-dependent interaction
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
17
Inelastic scenario
q q
DM DM*DMq ! DM? q
• If the splitting is positive the DM scatters into an heavier state: kinematic condition implies that the scatter occurs most probably with heavy nuclei (hence more sensitive to heavy WIMPs)
• If the splitting is negative exothermic Dark Matter, it decays into a lighter states and light target are favoured
Tucker-Smith, Weiner ’01
�m ⌘ (mDM? �mDM) = �
� ⇠ O(100 keV)
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
18
SI elastic scattering scenario CDSM-Si
lnLCDMSSi =
"3X
i=1
P1(S +B)
#+
2
4X
j
P0(S +B)
3
5+ lnLbck
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
1D marginalized posterior PDF for all parameters:
data from CDMS-Si collaboration arXiv:1304.4279
19
Combined fit more details
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
qNa0.2 0.3 0.4 0.5 0.60
0.2
0.4
0.6
0.8
1combined fit pushes the quenching factor and the local standard at rest at corner values (1D pdf not flat anymore but peaked at qNa=0.6)
Combined (elastic SI)
Single experiments
20
Combined fit more details
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
δ (keV)
Post
erio
r PD
F
0 50 100 150 2000
0.2
0.4
0.6
0.8
1
fn/fp
Post
erio
r PD
F
−2 −1.5 −1 −0.5 0 0.5 10
0.2
0.4
0.6
0.8
1
log10(mDM) (GeV)
log 10
(σnSI
) (cm
2 )
NFW density profile
Combined set of data
0 0.5 1 1.5−42
−41.5
−41
−40.5
−40
−39.5
−39
−38.5
−38
−37.5
−37
21
DAMA and CoGeNT, combined fit: hidden directions behavior
- the larger qNa the smaller the WIMP mass- low mass region is independent on qI
- similar behavior for the DM density at the sun position- less sensitive to the escape velocity value
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
XENON100
- S = 2 (seen events), likelihood follows a Poisson distribution - B = 1. +- 0.8- Total exposure 2323.7 kg days
Aprile et al. arXiv:1104.2549XENON100 collaboration, arXiv:1207.5988
- Scintillation efficiency is a systematic of the experimental set-up- treated as nuisance parameter with truncated gaussian prior and marginalized over
22 C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
XENON100
23
conversion between keVnr and PE
All the likelihoods are normalized such that if the background matches exactly the number of observed events
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
24
CoGeNT 2011Germanium cryogenic detectordetector mass 0.33 kglive time 442 daystotal exposure 145.86 kg days
- Data analysis and binning follow arXiv:1106.0650 [astro-ph.CO]- Radioactive peaks subtracted as prescribed by the collaboration- Analysis of the total rate with a background (27 bins)- Analysis of the modulated rate without background in 3 energy bins- All data are corrected by the efficiency factor, ranging from 0.7 to 0.82
Total rate : 27 bins of width 0.1 keVee energy range 0.5- 3.2 keVee
Modulated rate:
3 nuisance parameters for the non modulating background
quenching factor:
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
CoGeNT
DE1 = H0.5-0.9L keVee
0 100 200 300 400 50020
30
40
50
60
t HdaysL
Countsê3
0days
DE2 = H0.9-3.0L keVee
0 100 200 300 400 50020
40
60
80
100
t HdaysL
Countsê3
0days
DE3 = H3.0-4.5L keVee
0 100 200 300 400 50010
20
30
40
50
60
t HdaysL
Countsê3
0days
2.3�Modulation: from to 1.6�
Aalseth et al. arXiv:1106.0650
CA, J.Hamann, R.Trotta & Y.Wong arXiv:1111.3238 [hep-ph];
Ge detector, 146 kg days Very low threshold:0.4 keVee = 2.7 keV
25
Gaussian likelihood
• Background1. does not modulate, included only for the total rate2. constant + exponential background (mimic surface events)3. 3 nuisance parameters
• Radioactive peaks subtracted
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
CoGeNT 2011
26
Data analysisRadioactive peaks
arXiv:1106.0650
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
CRESST-II Angloher et al., arXiv:1109.0702evidence at 4 sigma
- 8 detector modules made by CaWO4 crystals (multi-target detectors)- scintillation + ionization to disentangle background
(e, n, alpha, decays of Pb isotopes)- exposure of 730 kg days- S = 67 events (background can account only for 65% of S)
light elements are more sensitive to light particles and viceversa!
27
Likelihood Poisson distribution1. Total counts in each module2. Global spectral information3. Background as nuisance parameters: 3 nuisance parameters
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
CRESST-II
28
Angloher et al., arXiv:1109.0702
- 8 detector module made by CaWO4 crystals (multi-target detectors)- scintillation + ionization to disentangle background (e, n, alpha, decays of Pb isotopes)- exposure of 730 kg days- S = 67 events (background can account only for 65% of N)
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
(In)Compatibility test
29
• We assume that:1. The fixed set is:2. The result to be tested is the number of events seen in the XENON100 detector
We find that elastic and isospin violating models are incompatible at 2 sigma level, while inelastic scattering is within 1 sigma
Inelastic
Isospin violating
Elastic with Nmax=60
Elastic wit Nmax=100
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
30
CoGeNT and combined fit disfavour inelastic scattering because the excess is in the low energy region and it prefers light WIMP masses
Which is the best model that accounts for the excessat low WIMP mass?
Elastic, inelastic and isospin violating models are nested models:Inelastic reduces to elastic for Isospin violating reduces to elastic for fn/fp = 1
� = 0
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
31
Parameter inference 3D: inelastic case
0 0.5 1 1.5 2 2.5 3−44
−43
−42
−41
−40
−39
−38
−37
−36
δ (keV)
log10(mDM) (GeV)
log 10
(σnSI
) (cm
2 )
0
2
4
6
8
10
12
14
16
18
20
0 0.5 1 1.5 2 2.5 3−44
−43
−42
−41
−40
−39
−38
−37
−36
δ (keV)
log10(mDM) (GeV)
log 10
(σn) (
cm2 )
0
20
40
60
80
100
120
140
160
180
200
0 0.5 1 1.5 2 2.5 3−44
−43
−42
−41
−40
−39
−38
−37
−36
δ (keV)
log10(mDM) (GeV)
log 10
(σn) (
cm2 )
0
20
40
60
80
100
120
140
160
180
200
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
32
Parameter inference 3D: isospin violating case
0 1 2 3−46
−44
−42
−40
−38
−36
fn/fp
log10(mDM) (GeV)
log 10
(σnSI
) (cm
2 )
−2
−1.5
−1
−0.5
0
0.5
1
0 1 2 3−44
−43
−42
−41
−40
−39
−38
−37
−36
fn/fp
log10(mDM) (GeV)
log 10
(σn) (
cm2 )
−2
−1.5
−1
−0.5
0
0.5
1
0 1 2 3−44
−43
−42
−41
−40
−39
−38
−37
−36
fn/fp
log10(mDM) (GeV)
log 10
(σn) (
cm2 )
−2
−1.5
−1
−0.5
0
0.5
1
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
33
Construction of DM velocity distribution (1)
DD depends on the distribution function (DF) at the sun position arising from the WIMPs phase-space distribution
• f(v) is a function of the gravitational potential (including baryon contribution)• f(v) is a function of the DM density profile
• DF obtained inverting the above equation • Symmetries assumed: density profile spherically symmetric and f(v) isotropic -> DF only function of the energy
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
34
Likelihood for astrophysical observables (nuisance parameters for ALL EXP)
The profiles mostly differ near the galactic center, at the sun position they give similar behavior for f(v)In what follow only shown comparison between NFW and SMH
Rü
0.001 0.01 0.1 1 10 1000.001
0.1
10
1000
r HkpcL
r DMHGe
Vêcm
3 LSpherically symmetric DM density profiles :
• NFW• Einasto• Cored Isothermal• Burkert
�DM = �DM(cvir,Mvir)
Construction of DM velocity distribution (2)
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
35
Sensitivity analysis
lnB2a = �1.06
• lnB of 1a:2a is now 3.11 instead of 5.21, still moderate evidence
• Results are robust from a Bayesian point of view!
For nested models with parameter priors separable the Savage Dickey density ratio (SDDR) gives an analytical estimate of the effect on lnB changing the width of the prior
-0.5 0.0 0.5 1.0 1.50.0
0.5
1.0
1.5
2.0
2.5
Prior width
pHq»M1Lno
rmalized
EXAMPLE
marginal posterior pdfs, computed at fixed value of the parameters
marginal normalized prior densitycomputed at fixed value of
36
Velocity distribution from DM density profile
Assuming equilibrium between gravitational force and pressure:
Eddigton formula for spherically symmetric DM density profiles that lead to isotropic f(v)
Poisson equation for the gravitational potential including contribution from the bulge and disk:
The velocity distribution is translated to the reference frame of the Earth:
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
Theoretical predictions for elastic spin-independent scattering off nucleus
Differential rate
Modulated rate
38 C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
Annual ModulationDrukier, Freese and Spergel ’86, Freese, Frieman and Gould ’88
In the Earth’s rest frame the DM velocity distribution acquires a time dependence, which follows a sinusoidal behavior
Signature of WIMP recoil in the detector
v2 = |�v⇥ + �v�|2
v� = |�v⇥ + �v���,rot|
= v⇥ + v⇤⇤�,rot cos � cos[2⇥(t� t0)/T ]
� = 60�
effect of O(10%)
Bernabei et al. arXiv:1002.1028
39 C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
41
Comparison between 5 phenomenological models that describe a sinusoidal modulation:
Is there evidence for DM modulation in CoGeNT?
DE1 = H0.5-0.9L keVee
0 100 200 300 400 50020
30
40
50
60
t HdaysL
Countsê3
0days
DE2 = H0.9-3.0L keVee
0 100 200 300 400 50020
40
60
80
100
t HdaysL
Countsê3
0days
DE3 = H3.0-4.5L keVee
0 100 200 300 400 50010
20
30
40
50
60
t HdaysLCountsê3
0days
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
42
-6 -4 -2 0 2 4 6ln(B)
0
1a
1b
2a
2bΔE1ΔE2ΔE3All ΔE
stron
g ev
iden
ce fo
r M0
mod
erat
e ev
iden
ce fo
r M0
wea
k ev
iden
ce fo
r M0
wea
k ev
iden
ce a
gain
st M
0
mod
erat
e ev
iden
ce a
gain
st M
0
stron
g ev
iden
ce a
gain
st M
0
-2.16
-1.88
0.88
0.57
1.49
-0.013
-0.98
0.61
-2.80
0.0048
-1.89
2.05
-3.16
1.50
-4.25
Non-DM T free
Non-DMannual
Consistent DM
Pheno-DM
Nomodulation
Bayes factor: results for model comparison
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
43
Classical p-valuesprobability of obtaining more extreme data than observed assuming the null hypothesis is correct and NOT probability for hypothesis
test statistics for nested models if1. additional dof distributed as a gaussian2. unbounded likelihood3. all additional dof identifiable under the null
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
43
Classical p-valuesprobability of obtaining more extreme data than observed assuming the null hypothesis is correct and NOT probability for hypothesis
test statistics for nested models if1. additional dof distributed as a gaussian2. unbounded likelihood3. all additional dof identifiable under the null
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
43
Classical p-valuesprobability of obtaining more extreme data than observed assuming the null hypothesis is correct and NOT probability for hypothesis
test statistics for nested models if1. additional dof distributed as a gaussian2. unbounded likelihood3. all additional dof identifiable under the nullx
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
43
Classical p-valuesprobability of obtaining more extreme data than observed assuming the null hypothesis is correct and NOT probability for hypothesis
Chernoff’s theorem
� =N�
i=0
2��
��
i
�p(�2
i > ��2e�)
test statistics for nested models if1. additional dof distributed as a gaussian2. unbounded likelihood3. all additional dof identifiable under the nullx
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
43
Classical p-valuesprobability of obtaining more extreme data than observed assuming the null hypothesis is correct and NOT probability for hypothesis
Chernoff’s theorem
� =N�
i=0
2��
��
i
�p(�2
i > ��2e�)
test statistics for nested models if1. additional dof distributed as a gaussian2. unbounded likelihood3. all additional dof identifiable under the nullx
2.3�
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
43
Classical p-valuesprobability of obtaining more extreme data than observed assuming the null hypothesis is correct and NOT probability for hypothesis
Chernoff’s theorem
� =N�
i=0
2��
��
i
�p(�2
i > ��2e�)
test statistics for nested models if1. additional dof distributed as a gaussian2. unbounded likelihood3. all additional dof identifiable under the nullx
1.6�
2.3�
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
43
Classical p-valuesprobability of obtaining more extreme data than observed assuming the null hypothesis is correct and NOT probability for hypothesis
Chernoff’s theorem
� =N�
i=0
2��
��
i
�p(�2
i > ��2e�)
test statistics for nested models if1. additional dof distributed as a gaussian2. unbounded likelihood3. all additional dof identifiable under the nullx
1.6�
2.3�
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
43
Classical p-valuesprobability of obtaining more extreme data than observed assuming the null hypothesis is correct and NOT probability for hypothesis
Chernoff’s theorem
� =N�
i=0
2��
��
i
�p(�2
i > ��2e�)
test statistics for nested models if1. additional dof distributed as a gaussian2. unbounded likelihood3. all additional dof identifiable under the nullxx
1.6�
2.3�
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
43
Classical p-valuesprobability of obtaining more extreme data than observed assuming the null hypothesis is correct and NOT probability for hypothesis
Chernoff’s theorem
� =N�
i=0
2��
��
i
�p(�2
i > ��2e�)
test statistics for nested models if1. additional dof distributed as a gaussian2. unbounded likelihood3. all additional dof identifiable under the nullxx
Rely on Monte Carlo simulation formapping the t statistic into p-values
1.6�
2.3�
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 201344
Similar behavior for the All bin case: the inference is driven by bin 2
Parameter inference: amplitude of
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 201344
Similar behavior for the All bin case: the inference is driven by bin 2
Parameter inference: amplitude of
45
Parameter inference: phase and period
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
46
Locally anisotropic DM velocity Ellipsoidal, triaxial DM halo model gives rise to a triaxial gaussian velocity distribution:
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
47
Background CoGeNTPriors on the fractional modulated amplitude predicted from configurations of DM mass and sigma that account for the CoGeNT total rate R(t) = S(t) + B
Background:1. does not modulate, included only for the total rate2. constant + exponential background (mimic surface events)
J. Collar talk @ TAUP 2011.
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
48
Model 1b: consistent DM Priors on the fractional modulated amplitude predicted from configurations of DM mass and sigma that account for the CoGeNT total rate R(t) = S(t) + B
Sm =R(tmax)�R(tmin)R(tmax) + R(tmin)
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013
49
Annual modulation in CoGeNT and in CDMS collaboration, Z. Amhed et al., arXiv:1203.1309 [astro-ph.CO]
• 214 kg days exposure•requirement that event scatter only once in the detector• 5 keVnr as threshold• No evidence for annual modulation
Amplitudes larger than are excluded at
C. Arina (IAP, Paris & GRAPPA Institute, UvA) - PASCOS 2013